drivers/net/bonding/bond_alb.c

   1 /*
   2  * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the GNU General Public License as published by the
   6  * Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful, but
  10  * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  11  * or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along
  15  * with this program; if not, see <http://www.gnu.org/licenses/>.
  16  *
  17  * The full GNU General Public License is included in this distribution in the
  18  * file called LICENSE.
  19  *
  20  */
  21
  22 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  23
  24 #include <linux/skbuff.h>
  25 #include <linux/netdevice.h>
  26 #include <linux/etherdevice.h>
  27 #include <linux/pkt_sched.h>
  28 #include <linux/spinlock.h>
  29 #include <linux/slab.h>
  30 #include <linux/timer.h>
  31 #include <linux/ip.h>
  32 #include <linux/ipv6.h>
  33 #include <linux/if_arp.h>
  34 #include <linux/if_ether.h>
  35 #include <linux/if_bonding.h>
  36 #include <linux/if_vlan.h>
  37 #include <linux/in.h>
  38 #include <net/ipx.h>
  39 #include <net/arp.h>
  40 #include <net/ipv6.h>
  41 #include <asm/byteorder.h>
  42 #include "bonding.h"
  43 #include "bond_alb.h"
  44
  45
  46
  47 #ifndef __long_aligned
  48 #define __long_aligned __attribute__((aligned((sizeof(long)))))
  49 #endif
  50 static const u8 mac_bcast[ETH_ALEN] __long_aligned = {
  51         0xff, 0xff, 0xff, 0xff, 0xff, 0xff
  52 };
  53 static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = {
  54         0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  55 };
  56 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  57
  58 #pragma pack(1)
  59 struct learning_pkt {
  60         u8 mac_dst[ETH_ALEN];
  61         u8 mac_src[ETH_ALEN];
  62         __be16 type;
  63         u8 padding[ETH_ZLEN - ETH_HLEN];
  64 };
  65
  66 struct arp_pkt {
  67         __be16  hw_addr_space;
  68         __be16  prot_addr_space;
  69         u8      hw_addr_len;
  70         u8      prot_addr_len;
  71         __be16  op_code;
  72         u8      mac_src[ETH_ALEN];      /* sender hardware address */
  73         __be32  ip_src;                 /* sender IP address */
  74         u8      mac_dst[ETH_ALEN];      /* target hardware address */
  75         __be32  ip_dst;                 /* target IP address */
  76 };
  77 #pragma pack()
  78
  79 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
  80 {
  81         return (struct arp_pkt *)skb_network_header(skb);
  82 }
  83
  84 /* Forward declaration */
  85 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
  86 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
  87 static void rlb_src_unlink(struct bonding *bond, u32 index);
  88 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
  89                          u32 ip_dst_hash);
  90
  91 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
  92 {
  93         int i;
  94         u8 hash = 0;
  95
  96         for (i = 0; i < hash_size; i++)
  97                 hash ^= hash_start[i];
  98
  99         return hash;
 100 }
 101
 102 /*********************** tlb specific functions ***************************/
 103
 104 static inline void _lock_tx_hashtbl_bh(struct bonding *bond)
 105 {
 106         spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 107 }
 108
 109 static inline void _unlock_tx_hashtbl_bh(struct bonding *bond)
 110 {
 111         spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 112 }
 113
 114 static inline void _lock_tx_hashtbl(struct bonding *bond)
 115 {
 116         spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 117 }
 118
 119 static inline void _unlock_tx_hashtbl(struct bonding *bond)
 120 {
 121         spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 122 }
 123
 124 /* Caller must hold tx_hashtbl lock */
 125 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 126 {
 127         if (save_load) {
 128                 entry->load_history = 1 + entry->tx_bytes /
 129                                       BOND_TLB_REBALANCE_INTERVAL;
 130                 entry->tx_bytes = 0;
 131         }
 132
 133         entry->tx_slave = NULL;
 134         entry->next = TLB_NULL_INDEX;
 135         entry->prev = TLB_NULL_INDEX;
 136 }
 137
 138 static inline void tlb_init_slave(struct slave *slave)
 139 {
 140         SLAVE_TLB_INFO(slave).load = 0;
 141         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 142 }
 143
 144 /* Caller must hold bond lock for read, BH disabled */
 145 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
 146                          int save_load)
 147 {
 148         struct tlb_client_info *tx_hash_table;
 149         u32 index;
 150
 151         /* clear slave from tx_hashtbl */
 152         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 153
 154         /* skip this if we've already freed the tx hash table */
 155         if (tx_hash_table) {
 156                 index = SLAVE_TLB_INFO(slave).head;
 157                 while (index != TLB_NULL_INDEX) {
 158                         u32 next_index = tx_hash_table[index].next;
 159                         tlb_init_table_entry(&tx_hash_table[index], save_load);
 160                         index = next_index;
 161                 }
 162         }
 163
 164         tlb_init_slave(slave);
 165 }
 166
 167 /* Caller must hold bond lock for read */
 168 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
 169                          int save_load)
 170 {
 171         _lock_tx_hashtbl_bh(bond);
 172         __tlb_clear_slave(bond, slave, save_load);
 173         _unlock_tx_hashtbl_bh(bond);
 174 }
 175
 176 /* Must be called before starting the monitor timer */
 177 static int tlb_initialize(struct bonding *bond)
 178 {
 179         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 180         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 181         struct tlb_client_info *new_hashtbl;
 182         int i;
 183
 184         new_hashtbl = kzalloc(size, GFP_KERNEL);
 185         if (!new_hashtbl)
 186                 return -1;
 187
 188         _lock_tx_hashtbl_bh(bond);
 189
 190         bond_info->tx_hashtbl = new_hashtbl;
 191
 192         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
 193                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
 194
 195         _unlock_tx_hashtbl_bh(bond);
 196
 197         return 0;
 198 }
 199
 200 /* Must be called only after all slaves have been released */
 201 static void tlb_deinitialize(struct bonding *bond)
 202 {
 203         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 204
 205         _lock_tx_hashtbl_bh(bond);
 206
 207         kfree(bond_info->tx_hashtbl);
 208         bond_info->tx_hashtbl = NULL;
 209
 210         _unlock_tx_hashtbl_bh(bond);
 211 }
 212
 213 static long long compute_gap(struct slave *slave)
 214 {
 215         return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
 216                (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 217 }
 218
 219 /* Caller must hold bond lock for read */
 220 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 221 {
 222         struct slave *slave, *least_loaded;
 223         struct list_head *iter;
 224         long long max_gap;
 225
 226         least_loaded = NULL;
 227         max_gap = LLONG_MIN;
 228
 229         /* Find the slave with the largest gap */
 230         bond_for_each_slave_rcu(bond, slave, iter) {
 231                 if (SLAVE_IS_OK(slave)) {
 232                         long long gap = compute_gap(slave);
 233
 234                         if (max_gap < gap) {
 235                                 least_loaded = slave;
 236                                 max_gap = gap;
 237                         }
 238                 }
 239         }
 240
 241         return least_loaded;
 242 }
 243
 244 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
 245                                                 u32 skb_len)
 246 {
 247         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 248         struct tlb_client_info *hash_table;
 249         struct slave *assigned_slave;
 250
 251         hash_table = bond_info->tx_hashtbl;
 252         assigned_slave = hash_table[hash_index].tx_slave;
 253         if (!assigned_slave) {
 254                 assigned_slave = tlb_get_least_loaded_slave(bond);
 255
 256                 if (assigned_slave) {
 257                         struct tlb_slave_info *slave_info =
 258                                 &(SLAVE_TLB_INFO(assigned_slave));
 259                         u32 next_index = slave_info->head;
 260
 261                         hash_table[hash_index].tx_slave = assigned_slave;
 262                         hash_table[hash_index].next = next_index;
 263                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 264
 265                         if (next_index != TLB_NULL_INDEX)
 266                                 hash_table[next_index].prev = hash_index;
 267
 268                         slave_info->head = hash_index;
 269                         slave_info->load +=
 270                                 hash_table[hash_index].load_history;
 271                 }
 272         }
 273
 274         if (assigned_slave)
 275                 hash_table[hash_index].tx_bytes += skb_len;
 276
 277         return assigned_slave;
 278 }
 279
 280 /* Caller must hold bond lock for read */
 281 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
 282                                         u32 skb_len)
 283 {
 284         struct slave *tx_slave;
 285         /*
 286          * We don't need to disable softirq here, becase
 287          * tlb_choose_channel() is only called by bond_alb_xmit()
 288          * which already has softirq disabled.
 289          */
 290         _lock_tx_hashtbl(bond);
 291         tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
 292         _unlock_tx_hashtbl(bond);
 293         return tx_slave;
 294 }
 295
 296 /*********************** rlb specific functions ***************************/
 297 static inline void _lock_rx_hashtbl_bh(struct bonding *bond)
 298 {
 299         spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 300 }
 301
 302 static inline void _unlock_rx_hashtbl_bh(struct bonding *bond)
 303 {
 304         spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 305 }
 306
 307 static inline void _lock_rx_hashtbl(struct bonding *bond)
 308 {
 309         spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 310 }
 311
 312 static inline void _unlock_rx_hashtbl(struct bonding *bond)
 313 {
 314         spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 315 }
 316
 317 /* when an ARP REPLY is received from a client update its info
 318  * in the rx_hashtbl
 319  */
 320 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 321 {
 322         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 323         struct rlb_client_info *client_info;
 324         u32 hash_index;
 325
 326         _lock_rx_hashtbl_bh(bond);
 327
 328         hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 329         client_info = &(bond_info->rx_hashtbl[hash_index]);
 330
 331         if ((client_info->assigned) &&
 332             (client_info->ip_src == arp->ip_dst) &&
 333             (client_info->ip_dst == arp->ip_src) &&
 334             (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
 335                 /* update the clients MAC address */
 336                 ether_addr_copy(client_info->mac_dst, arp->mac_src);
 337                 client_info->ntt = 1;
 338                 bond_info->rx_ntt = 1;
 339         }
 340
 341         _unlock_rx_hashtbl_bh(bond);
 342 }
 343
 344 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
 345                         struct slave *slave)
 346 {
 347         struct arp_pkt *arp, _arp;
 348
 349         if (skb->protocol != cpu_to_be16(ETH_P_ARP))
 350                 goto out;
 351
 352         arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
 353         if (!arp)
 354                 goto out;
 355
 356         /* We received an ARP from arp->ip_src.
 357          * We might have used this IP address previously (on the bonding host
 358          * itself or on a system that is bridged together with the bond).
 359          * However, if arp->mac_src is different than what is stored in
 360          * rx_hashtbl, some other host is now using the IP and we must prevent
 361          * sending out client updates with this IP address and the old MAC
 362          * address.
 363          * Clean up all hash table entries that have this address as ip_src but
 364          * have a different mac_src.
 365          */
 366         rlb_purge_src_ip(bond, arp);
 367
 368         if (arp->op_code == htons(ARPOP_REPLY)) {
 369                 /* update rx hash table for this ARP */
 370                 rlb_update_entry_from_arp(bond, arp);
 371                 pr_debug("Server received an ARP Reply from client\n");
 372         }
 373 out:
 374         return RX_HANDLER_ANOTHER;
 375 }
 376
 377 /* Caller must hold bond lock for read */
 378 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 379 {
 380         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 381         struct slave *before = NULL, *rx_slave = NULL, *slave;
 382         struct list_head *iter;
 383         bool found = false;
 384
 385         bond_for_each_slave(bond, slave, iter) {
 386                 if (!SLAVE_IS_OK(slave))
 387                         continue;
 388                 if (!found) {
 389                         if (!before || before->speed < slave->speed)
 390                                 before = slave;
 391                 } else {
 392                         if (!rx_slave || rx_slave->speed < slave->speed)
 393                                 rx_slave = slave;
 394                 }
 395                 if (slave == bond_info->rx_slave)
 396                         found = true;
 397         }
 398         /* we didn't find anything after the current or we have something
 399          * better before and up to the current slave
 400          */
 401         if (!rx_slave || (before && rx_slave->speed < before->speed))
 402                 rx_slave = before;
 403
 404         if (rx_slave)
 405                 bond_info->rx_slave = rx_slave;
 406
 407         return rx_slave;
 408 }
 409
 410 /* Caller must hold rcu_read_lock() for read */
 411 static struct slave *__rlb_next_rx_slave(struct bonding *bond)
 412 {
 413         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 414         struct slave *before = NULL, *rx_slave = NULL, *slave;
 415         struct list_head *iter;
 416         bool found = false;
 417
 418         bond_for_each_slave_rcu(bond, slave, iter) {
 419                 if (!SLAVE_IS_OK(slave))
 420                         continue;
 421                 if (!found) {
 422                         if (!before || before->speed < slave->speed)
 423                                 before = slave;
 424                 } else {
 425                         if (!rx_slave || rx_slave->speed < slave->speed)
 426                                 rx_slave = slave;
 427                 }
 428                 if (slave == bond_info->rx_slave)
 429                         found = true;
 430         }
 431         /* we didn't find anything after the current or we have something
 432          * better before and up to the current slave
 433          */
 434         if (!rx_slave || (before && rx_slave->speed < before->speed))
 435                 rx_slave = before;
 436
 437         if (rx_slave)
 438                 bond_info->rx_slave = rx_slave;
 439
 440         return rx_slave;
 441 }
 442
 443 /* teach the switch the mac of a disabled slave
 444  * on the primary for fault tolerance
 445  *
 446  * Caller must hold bond->curr_slave_lock for write or bond lock for write
 447  */
 448 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 449 {
 450         if (!bond->curr_active_slave)
 451                 return;
 452
 453         if (!bond->alb_info.primary_is_promisc) {
 454                 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
 455                         bond->alb_info.primary_is_promisc = 1;
 456                 else
 457                         bond->alb_info.primary_is_promisc = 0;
 458         }
 459
 460         bond->alb_info.rlb_promisc_timeout_counter = 0;
 461
 462         alb_send_learning_packets(bond->curr_active_slave, addr);
 463 }
 464
 465 /* slave being removed should not be active at this point
 466  *
 467  * Caller must hold rtnl.
 468  */
 469 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 470 {
 471         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 472         struct rlb_client_info *rx_hash_table;
 473         u32 index, next_index;
 474
 475         /* clear slave from rx_hashtbl */
 476         _lock_rx_hashtbl_bh(bond);
 477
 478         rx_hash_table = bond_info->rx_hashtbl;
 479         index = bond_info->rx_hashtbl_used_head;
 480         for (; index != RLB_NULL_INDEX; index = next_index) {
 481                 next_index = rx_hash_table[index].used_next;
 482                 if (rx_hash_table[index].slave == slave) {
 483                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 484
 485                         if (assigned_slave) {
 486                                 rx_hash_table[index].slave = assigned_slave;
 487                                 if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
 488                                                              mac_bcast)) {
 489                                         bond_info->rx_hashtbl[index].ntt = 1;
 490                                         bond_info->rx_ntt = 1;
 491                                         /* A slave has been removed from the
 492                                          * table because it is either disabled
 493                                          * or being released. We must retry the
 494                                          * update to avoid clients from not
 495                                          * being updated & disconnecting when
 496                                          * there is stress
 497                                          */
 498                                         bond_info->rlb_update_retry_counter =
 499                                                 RLB_UPDATE_RETRY;
 500                                 }
 501                         } else {  /* there is no active slave */
 502                                 rx_hash_table[index].slave = NULL;
 503                         }
 504                 }
 505         }
 506
 507         _unlock_rx_hashtbl_bh(bond);
 508
 509         write_lock_bh(&bond->curr_slave_lock);
 510
 511         if (slave != bond->curr_active_slave)
 512                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 513
 514         write_unlock_bh(&bond->curr_slave_lock);
 515 }
 516
 517 static void rlb_update_client(struct rlb_client_info *client_info)
 518 {
 519         int i;
 520
 521         if (!client_info->slave)
 522                 return;
 523
 524         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 525                 struct sk_buff *skb;
 526
 527                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 528                                  client_info->ip_dst,
 529                                  client_info->slave->dev,
 530                                  client_info->ip_src,
 531                                  client_info->mac_dst,
 532                                  client_info->slave->dev->dev_addr,
 533                                  client_info->mac_dst);
 534                 if (!skb) {
 535                         pr_err("%s: Error: failed to create an ARP packet\n",
 536                                client_info->slave->bond->dev->name);
 537                         continue;
 538                 }
 539
 540                 skb->dev = client_info->slave->dev;
 541
 542                 if (client_info->vlan_id) {
 543                         skb = vlan_put_tag(skb, htons(ETH_P_8021Q), client_info->vlan_id);
 544                         if (!skb) {
 545                                 pr_err("%s: Error: failed to insert VLAN tag\n",
 546                                        client_info->slave->bond->dev->name);
 547                                 continue;
 548                         }
 549                 }
 550
 551                 arp_xmit(skb);
 552         }
 553 }
 554
 555 /* sends ARP REPLIES that update the clients that need updating */
 556 static void rlb_update_rx_clients(struct bonding *bond)
 557 {
 558         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 559         struct rlb_client_info *client_info;
 560         u32 hash_index;
 561
 562         _lock_rx_hashtbl_bh(bond);
 563
 564         hash_index = bond_info->rx_hashtbl_used_head;
 565         for (; hash_index != RLB_NULL_INDEX;
 566              hash_index = client_info->used_next) {
 567                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 568                 if (client_info->ntt) {
 569                         rlb_update_client(client_info);
 570                         if (bond_info->rlb_update_retry_counter == 0)
 571                                 client_info->ntt = 0;
 572                 }
 573         }
 574
 575         /* do not update the entries again until this counter is zero so that
 576          * not to confuse the clients.
 577          */
 578         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 579
 580         _unlock_rx_hashtbl_bh(bond);
 581 }
 582
 583 /* The slave was assigned a new mac address - update the clients */
 584 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 585 {
 586         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 587         struct rlb_client_info *client_info;
 588         int ntt = 0;
 589         u32 hash_index;
 590
 591         _lock_rx_hashtbl_bh(bond);
 592
 593         hash_index = bond_info->rx_hashtbl_used_head;
 594         for (; hash_index != RLB_NULL_INDEX;
 595              hash_index = client_info->used_next) {
 596                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 597
 598                 if ((client_info->slave == slave) &&
 599                     !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 600                         client_info->ntt = 1;
 601                         ntt = 1;
 602                 }
 603         }
 604
 605         /* update the team's flag only after the whole iteration */
 606         if (ntt) {
 607                 bond_info->rx_ntt = 1;
 608                 /* fasten the change */
 609                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 610         }
 611
 612         _unlock_rx_hashtbl_bh(bond);
 613 }
 614
 615 /* mark all clients using src_ip to be updated */
 616 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 617 {
 618         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 619         struct rlb_client_info *client_info;
 620         u32 hash_index;
 621
 622         _lock_rx_hashtbl(bond);
 623
 624         hash_index = bond_info->rx_hashtbl_used_head;
 625         for (; hash_index != RLB_NULL_INDEX;
 626              hash_index = client_info->used_next) {
 627                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 628
 629                 if (!client_info->slave) {
 630                         pr_err("%s: Error: found a client with no channel in the client's hash table\n",
 631                                bond->dev->name);
 632                         continue;
 633                 }
 634                 /*update all clients using this src_ip, that are not assigned
 635                  * to the team's address (curr_active_slave) and have a known
 636                  * unicast mac address.
 637                  */
 638                 if ((client_info->ip_src == src_ip) &&
 639                     !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
 640                                              bond->dev->dev_addr) &&
 641                     !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 642                         client_info->ntt = 1;
 643                         bond_info->rx_ntt = 1;
 644                 }
 645         }
 646
 647         _unlock_rx_hashtbl(bond);
 648 }
 649
 650 /* Caller must hold both bond and ptr locks for read */
 651 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 652 {
 653         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 654         struct arp_pkt *arp = arp_pkt(skb);
 655         struct slave *assigned_slave, *curr_active_slave;
 656         struct rlb_client_info *client_info;
 657         u32 hash_index = 0;
 658
 659         _lock_rx_hashtbl(bond);
 660
 661         curr_active_slave = rcu_dereference(bond->curr_active_slave);
 662
 663         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 664         client_info = &(bond_info->rx_hashtbl[hash_index]);
 665
 666         if (client_info->assigned) {
 667                 if ((client_info->ip_src == arp->ip_src) &&
 668                     (client_info->ip_dst == arp->ip_dst)) {
 669                         /* the entry is already assigned to this client */
 670                         if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
 671                                 /* update mac address from arp */
 672                                 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 673                         }
 674                         ether_addr_copy(client_info->mac_src, arp->mac_src);
 675
 676                         assigned_slave = client_info->slave;
 677                         if (assigned_slave) {
 678                                 _unlock_rx_hashtbl(bond);
 679                                 return assigned_slave;
 680                         }
 681                 } else {
 682                         /* the entry is already assigned to some other client,
 683                          * move the old client to primary (curr_active_slave) so
 684                          * that the new client can be assigned to this entry.
 685                          */
 686                         if (bond->curr_active_slave &&
 687                             client_info->slave != curr_active_slave) {
 688                                 client_info->slave = curr_active_slave;
 689                                 rlb_update_client(client_info);
 690                         }
 691                 }
 692         }
 693         /* assign a new slave */
 694         assigned_slave = __rlb_next_rx_slave(bond);
 695
 696         if (assigned_slave) {
 697                 if (!(client_info->assigned &&
 698                       client_info->ip_src == arp->ip_src)) {
 699                         /* ip_src is going to be updated,
 700                          * fix the src hash list
 701                          */
 702                         u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
 703                                                     sizeof(arp->ip_src));
 704                         rlb_src_unlink(bond, hash_index);
 705                         rlb_src_link(bond, hash_src, hash_index);
 706                 }
 707
 708                 client_info->ip_src = arp->ip_src;
 709                 client_info->ip_dst = arp->ip_dst;
 710                 /* arp->mac_dst is broadcast for arp reqeusts.
 711                  * will be updated with clients actual unicast mac address
 712                  * upon receiving an arp reply.
 713                  */
 714                 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 715                 ether_addr_copy(client_info->mac_src, arp->mac_src);
 716                 client_info->slave = assigned_slave;
 717
 718                 if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 719                         client_info->ntt = 1;
 720                         bond->alb_info.rx_ntt = 1;
 721                 } else {
 722                         client_info->ntt = 0;
 723                 }
 724
 725                 if (vlan_get_tag(skb, &client_info->vlan_id))
 726                         client_info->vlan_id = 0;
 727
 728                 if (!client_info->assigned) {
 729                         u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
 730                         bond_info->rx_hashtbl_used_head = hash_index;
 731                         client_info->used_next = prev_tbl_head;
 732                         if (prev_tbl_head != RLB_NULL_INDEX) {
 733                                 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
 734                                         hash_index;
 735                         }
 736                         client_info->assigned = 1;
 737                 }
 738         }
 739
 740         _unlock_rx_hashtbl(bond);
 741
 742         return assigned_slave;
 743 }
 744
 745 /* chooses (and returns) transmit channel for arp reply
 746  * does not choose channel for other arp types since they are
 747  * sent on the curr_active_slave
 748  */
 749 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 750 {
 751         struct arp_pkt *arp = arp_pkt(skb);
 752         struct slave *tx_slave = NULL;
 753
 754         /* Don't modify or load balance ARPs that do not originate locally
 755          * (e.g.,arrive via a bridge).
 756          */
 757         if (!bond_slave_has_mac_rcu(bond, arp->mac_src))
 758                 return NULL;
 759
 760         if (arp->op_code == htons(ARPOP_REPLY)) {
 761                 /* the arp must be sent on the selected
 762                 * rx channel
 763                 */
 764                 tx_slave = rlb_choose_channel(skb, bond);
 765                 if (tx_slave)
 766                         ether_addr_copy(arp->mac_src, tx_slave->dev->dev_addr);
 767                 pr_debug("Server sent ARP Reply packet\n");
 768         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 769                 /* Create an entry in the rx_hashtbl for this client as a
 770                  * place holder.
 771                  * When the arp reply is received the entry will be updated
 772                  * with the correct unicast address of the client.
 773                  */
 774                 rlb_choose_channel(skb, bond);
 775
 776                 /* The ARP reply packets must be delayed so that
 777                  * they can cancel out the influence of the ARP request.
 778                  */
 779                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 780
 781                 /* arp requests are broadcast and are sent on the primary
 782                  * the arp request will collapse all clients on the subnet to
 783                  * the primary slave. We must register these clients to be
 784                  * updated with their assigned mac.
 785                  */
 786                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 787                 pr_debug("Server sent ARP Request packet\n");
 788         }
 789
 790         return tx_slave;
 791 }
 792
 793 /* Caller must hold bond lock for read */
 794 static void rlb_rebalance(struct bonding *bond)
 795 {
 796         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 797         struct slave *assigned_slave;
 798         struct rlb_client_info *client_info;
 799         int ntt;
 800         u32 hash_index;
 801
 802         _lock_rx_hashtbl_bh(bond);
 803
 804         ntt = 0;
 805         hash_index = bond_info->rx_hashtbl_used_head;
 806         for (; hash_index != RLB_NULL_INDEX;
 807              hash_index = client_info->used_next) {
 808                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 809                 assigned_slave = __rlb_next_rx_slave(bond);
 810                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 811                         client_info->slave = assigned_slave;
 812                         client_info->ntt = 1;
 813                         ntt = 1;
 814                 }
 815         }
 816
 817         /* update the team's flag only after the whole iteration */
 818         if (ntt)
 819                 bond_info->rx_ntt = 1;
 820         _unlock_rx_hashtbl_bh(bond);
 821 }
 822
 823 /* Caller must hold rx_hashtbl lock */
 824 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
 825 {
 826         entry->used_next = RLB_NULL_INDEX;
 827         entry->used_prev = RLB_NULL_INDEX;
 828         entry->assigned = 0;
 829         entry->slave = NULL;
 830         entry->vlan_id = 0;
 831 }
 832 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
 833 {
 834         entry->src_first = RLB_NULL_INDEX;
 835         entry->src_prev = RLB_NULL_INDEX;
 836         entry->src_next = RLB_NULL_INDEX;
 837 }
 838
 839 static void rlb_init_table_entry(struct rlb_client_info *entry)
 840 {
 841         memset(entry, 0, sizeof(struct rlb_client_info));
 842         rlb_init_table_entry_dst(entry);
 843         rlb_init_table_entry_src(entry);
 844 }
 845
 846 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
 847 {
 848         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 849         u32 next_index = bond_info->rx_hashtbl[index].used_next;
 850         u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
 851
 852         if (index == bond_info->rx_hashtbl_used_head)
 853                 bond_info->rx_hashtbl_used_head = next_index;
 854         if (prev_index != RLB_NULL_INDEX)
 855                 bond_info->rx_hashtbl[prev_index].used_next = next_index;
 856         if (next_index != RLB_NULL_INDEX)
 857                 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
 858 }
 859
 860 /* unlink a rlb hash table entry from the src list */
 861 static void rlb_src_unlink(struct bonding *bond, u32 index)
 862 {
 863         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 864         u32 next_index = bond_info->rx_hashtbl[index].src_next;
 865         u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
 866
 867         bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
 868         bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
 869
 870         if (next_index != RLB_NULL_INDEX)
 871                 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
 872
 873         if (prev_index == RLB_NULL_INDEX)
 874                 return;
 875
 876         /* is prev_index pointing to the head of this list? */
 877         if (bond_info->rx_hashtbl[prev_index].src_first == index)
 878                 bond_info->rx_hashtbl[prev_index].src_first = next_index;
 879         else
 880                 bond_info->rx_hashtbl[prev_index].src_next = next_index;
 881
 882 }
 883
 884 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
 885 {
 886         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 887         struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 888
 889         rlb_delete_table_entry_dst(bond, index);
 890         rlb_init_table_entry_dst(entry);
 891
 892         rlb_src_unlink(bond, index);
 893 }
 894
 895 /* add the rx_hashtbl[ip_dst_hash] entry to the list
 896  * of entries with identical ip_src_hash
 897  */
 898 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
 899 {
 900         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 901         u32 next;
 902
 903         bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
 904         next = bond_info->rx_hashtbl[ip_src_hash].src_first;
 905         bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
 906         if (next != RLB_NULL_INDEX)
 907                 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
 908         bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
 909 }
 910
 911 /* deletes all rx_hashtbl entries with  arp->ip_src if their mac_src does
 912  * not match arp->mac_src */
 913 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
 914 {
 915         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 916         u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 917         u32 index;
 918
 919         _lock_rx_hashtbl_bh(bond);
 920
 921         index = bond_info->rx_hashtbl[ip_src_hash].src_first;
 922         while (index != RLB_NULL_INDEX) {
 923                 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 924                 u32 next_index = entry->src_next;
 925                 if (entry->ip_src == arp->ip_src &&
 926                     !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
 927                                 rlb_delete_table_entry(bond, index);
 928                 index = next_index;
 929         }
 930         _unlock_rx_hashtbl_bh(bond);
 931 }
 932
 933 static int rlb_initialize(struct bonding *bond)
 934 {
 935         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 936         struct rlb_client_info  *new_hashtbl;
 937         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 938         int i;
 939
 940         new_hashtbl = kmalloc(size, GFP_KERNEL);
 941         if (!new_hashtbl)
 942                 return -1;
 943
 944         _lock_rx_hashtbl_bh(bond);
 945
 946         bond_info->rx_hashtbl = new_hashtbl;
 947
 948         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 949
 950         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
 951                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 952
 953         _unlock_rx_hashtbl_bh(bond);
 954
 955         /* register to receive ARPs */
 956         bond->recv_probe = rlb_arp_recv;
 957
 958         return 0;
 959 }
 960
 961 static void rlb_deinitialize(struct bonding *bond)
 962 {
 963         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 964
 965         _lock_rx_hashtbl_bh(bond);
 966
 967         kfree(bond_info->rx_hashtbl);
 968         bond_info->rx_hashtbl = NULL;
 969         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 970
 971         _unlock_rx_hashtbl_bh(bond);
 972 }
 973
 974 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 975 {
 976         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 977         u32 curr_index;
 978
 979         _lock_rx_hashtbl_bh(bond);
 980
 981         curr_index = bond_info->rx_hashtbl_used_head;
 982         while (curr_index != RLB_NULL_INDEX) {
 983                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 984                 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
 985
 986                 if (curr->vlan_id == vlan_id)
 987                         rlb_delete_table_entry(bond, curr_index);
 988
 989                 curr_index = next_index;
 990         }
 991
 992         _unlock_rx_hashtbl_bh(bond);
 993 }
 994
 995 /*********************** tlb/rlb shared functions *********************/
 996
 997 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
 998                             u16 vid)
 999 {
1000         struct learning_pkt pkt;
1001         struct sk_buff *skb;
1002         int size = sizeof(struct learning_pkt);
1003         char *data;
1004
1005         memset(&pkt, 0, size);
1006         ether_addr_copy(pkt.mac_dst, mac_addr);
1007         ether_addr_copy(pkt.mac_src, mac_addr);
1008         pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
1009
1010         skb = dev_alloc_skb(size);
1011         if (!skb)
1012                 return;
1013
1014         data = skb_put(skb, size);
1015         memcpy(data, &pkt, size);
1016
1017         skb_reset_mac_header(skb);
1018         skb->network_header = skb->mac_header + ETH_HLEN;
1019         skb->protocol = pkt.type;
1020         skb->priority = TC_PRIO_CONTROL;
1021         skb->dev = slave->dev;
1022
1023         if (vid) {
1024                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vid);
1025                 if (!skb) {
1026                         pr_err("%s: Error: failed to insert VLAN tag\n",
1027                                slave->bond->dev->name);
1028                         return;
1029                 }
1030         }
1031
1032         dev_queue_xmit(skb);
1033 }
1034
1035
1036 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
1037 {
1038         struct bonding *bond = bond_get_bond_by_slave(slave);
1039         struct net_device *upper;
1040         struct list_head *iter;
1041
1042         /* send untagged */
1043         alb_send_lp_vid(slave, mac_addr, 0);
1044
1045         /* loop through vlans and send one packet for each */
1046         rcu_read_lock();
1047         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
1048                 if (upper->priv_flags & IFF_802_1Q_VLAN)
1049                         alb_send_lp_vid(slave, mac_addr,
1050                                         vlan_dev_vlan_id(upper));
1051         }
1052         rcu_read_unlock();
1053 }
1054
1055 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[])
1056 {
1057         struct net_device *dev = slave->dev;
1058         struct sockaddr s_addr;
1059
1060         if (slave->bond->params.mode == BOND_MODE_TLB) {
1061                 memcpy(dev->dev_addr, addr, dev->addr_len);
1062                 return 0;
1063         }
1064
1065         /* for rlb each slave must have a unique hw mac addresses so that */
1066         /* each slave will receive packets destined to a different mac */
1067         memcpy(s_addr.sa_data, addr, dev->addr_len);
1068         s_addr.sa_family = dev->type;
1069         if (dev_set_mac_address(dev, &s_addr)) {
1070                 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
1071                        "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1072                        slave->bond->dev->name, dev->name);
1073                 return -EOPNOTSUPP;
1074         }
1075         return 0;
1076 }
1077
1078 /*
1079  * Swap MAC addresses between two slaves.
1080  *
1081  * Called with RTNL held, and no other locks.
1082  *
1083  */
1084
1085 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1086 {
1087         u8 tmp_mac_addr[ETH_ALEN];
1088
1089         ether_addr_copy(tmp_mac_addr, slave1->dev->dev_addr);
1090         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr);
1091         alb_set_slave_mac_addr(slave2, tmp_mac_addr);
1092
1093 }
1094
1095 /*
1096  * Send learning packets after MAC address swap.
1097  *
1098  * Called with RTNL and no other locks
1099  */
1100 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1101                                 struct slave *slave2)
1102 {
1103         int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
1104         struct slave *disabled_slave = NULL;
1105
1106         ASSERT_RTNL();
1107
1108         /* fasten the change in the switch */
1109         if (SLAVE_IS_OK(slave1)) {
1110                 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
1111                 if (bond->alb_info.rlb_enabled) {
1112                         /* inform the clients that the mac address
1113                          * has changed
1114                          */
1115                         rlb_req_update_slave_clients(bond, slave1);
1116                 }
1117         } else {
1118                 disabled_slave = slave1;
1119         }
1120
1121         if (SLAVE_IS_OK(slave2)) {
1122                 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
1123                 if (bond->alb_info.rlb_enabled) {
1124                         /* inform the clients that the mac address
1125                          * has changed
1126                          */
1127                         rlb_req_update_slave_clients(bond, slave2);
1128                 }
1129         } else {
1130                 disabled_slave = slave2;
1131         }
1132
1133         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1134                 /* A disabled slave was assigned an active mac addr */
1135                 rlb_teach_disabled_mac_on_primary(bond,
1136                                                   disabled_slave->dev->dev_addr);
1137         }
1138 }
1139
1140 /**
1141  * alb_change_hw_addr_on_detach
1142  * @bond: bonding we're working on
1143  * @slave: the slave that was just detached
1144  *
1145  * We assume that @slave was already detached from the slave list.
1146  *
1147  * If @slave's permanent hw address is different both from its current
1148  * address and from @bond's address, then somewhere in the bond there's
1149  * a slave that has @slave's permanet address as its current address.
1150  * We'll make sure that that slave no longer uses @slave's permanent address.
1151  *
1152  * Caller must hold RTNL and no other locks
1153  */
1154 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1155 {
1156         int perm_curr_diff;
1157         int perm_bond_diff;
1158         struct slave *found_slave;
1159
1160         perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1161                                                   slave->dev->dev_addr);
1162         perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1163                                                   bond->dev->dev_addr);
1164
1165         if (perm_curr_diff && perm_bond_diff) {
1166                 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1167
1168                 if (found_slave) {
1169                         /* locking: needs RTNL and nothing else */
1170                         alb_swap_mac_addr(slave, found_slave);
1171                         alb_fasten_mac_swap(bond, slave, found_slave);
1172                 }
1173         }
1174 }
1175
1176 /**
1177  * alb_handle_addr_collision_on_attach
1178  * @bond: bonding we're working on
1179  * @slave: the slave that was just attached
1180  *
1181  * checks uniqueness of slave's mac address and handles the case the
1182  * new slave uses the bonds mac address.
1183  *
1184  * If the permanent hw address of @slave is @bond's hw address, we need to
1185  * find a different hw address to give @slave, that isn't in use by any other
1186  * slave in the bond. This address must be, of course, one of the permanent
1187  * addresses of the other slaves.
1188  *
1189  * We go over the slave list, and for each slave there we compare its
1190  * permanent hw address with the current address of all the other slaves.
1191  * If no match was found, then we've found a slave with a permanent address
1192  * that isn't used by any other slave in the bond, so we can assign it to
1193  * @slave.
1194  *
1195  * assumption: this function is called before @slave is attached to the
1196  *             bond slave list.
1197  */
1198 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1199 {
1200         struct slave *has_bond_addr = bond->curr_active_slave;
1201         struct slave *tmp_slave1, *free_mac_slave = NULL;
1202         struct list_head *iter;
1203
1204         if (!bond_has_slaves(bond)) {
1205                 /* this is the first slave */
1206                 return 0;
1207         }
1208
1209         /* if slave's mac address differs from bond's mac address
1210          * check uniqueness of slave's mac address against the other
1211          * slaves in the bond.
1212          */
1213         if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1214                 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1215                         return 0;
1216
1217                 /* Try setting slave mac to bond address and fall-through
1218                    to code handling that situation below... */
1219                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr);
1220         }
1221
1222         /* The slave's address is equal to the address of the bond.
1223          * Search for a spare address in the bond for this slave.
1224          */
1225         bond_for_each_slave(bond, tmp_slave1, iter) {
1226                 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1227                         /* no slave has tmp_slave1's perm addr
1228                          * as its curr addr
1229                          */
1230                         free_mac_slave = tmp_slave1;
1231                         break;
1232                 }
1233
1234                 if (!has_bond_addr) {
1235                         if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1236                                                     bond->dev->dev_addr)) {
1237
1238                                 has_bond_addr = tmp_slave1;
1239                         }
1240                 }
1241         }
1242
1243         if (free_mac_slave) {
1244                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr);
1245
1246                 pr_warn("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1247                         bond->dev->name, slave->dev->name,
1248                         free_mac_slave->dev->name);
1249
1250         } else if (has_bond_addr) {
1251                 pr_err("%s: Error: the hw address of slave %s is in use by the bond; couldn't find a slave with a free hw address to give it (this should not have happened)\n",
1252                        bond->dev->name, slave->dev->name);
1253                 return -EFAULT;
1254         }
1255
1256         return 0;
1257 }
1258
1259 /**
1260  * alb_set_mac_address
1261  * @bond:
1262  * @addr:
1263  *
1264  * In TLB mode all slaves are configured to the bond's hw address, but set
1265  * their dev_addr field to different addresses (based on their permanent hw
1266  * addresses).
1267  *
1268  * For each slave, this function sets the interface to the new address and then
1269  * changes its dev_addr field to its previous value.
1270  *
1271  * Unwinding assumes bond's mac address has not yet changed.
1272  */
1273 static int alb_set_mac_address(struct bonding *bond, void *addr)
1274 {
1275         struct slave *slave, *rollback_slave;
1276         struct list_head *iter;
1277         struct sockaddr sa;
1278         char tmp_addr[ETH_ALEN];
1279         int res;
1280
1281         if (bond->alb_info.rlb_enabled)
1282                 return 0;
1283
1284         bond_for_each_slave(bond, slave, iter) {
1285                 /* save net_device's current hw address */
1286                 ether_addr_copy(tmp_addr, slave->dev->dev_addr);
1287
1288                 res = dev_set_mac_address(slave->dev, addr);
1289
1290                 /* restore net_device's hw address */
1291                 ether_addr_copy(slave->dev->dev_addr, tmp_addr);
1292
1293                 if (res)
1294                         goto unwind;
1295         }
1296
1297         return 0;
1298
1299 unwind:
1300         memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1301         sa.sa_family = bond->dev->type;
1302
1303         /* unwind from head to the slave that failed */
1304         bond_for_each_slave(bond, rollback_slave, iter) {
1305                 if (rollback_slave == slave)
1306                         break;
1307                 ether_addr_copy(tmp_addr, rollback_slave->dev->dev_addr);
1308                 dev_set_mac_address(rollback_slave->dev, &sa);
1309                 ether_addr_copy(rollback_slave->dev->dev_addr, tmp_addr);
1310         }
1311
1312         return res;
1313 }
1314
1315 /************************ exported alb funcions ************************/
1316
1317 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1318 {
1319         int res;
1320
1321         res = tlb_initialize(bond);
1322         if (res)
1323                 return res;
1324
1325         if (rlb_enabled) {
1326                 bond->alb_info.rlb_enabled = 1;
1327                 /* initialize rlb */
1328                 res = rlb_initialize(bond);
1329                 if (res) {
1330                         tlb_deinitialize(bond);
1331                         return res;
1332                 }
1333         } else {
1334                 bond->alb_info.rlb_enabled = 0;
1335         }
1336
1337         return 0;
1338 }
1339
1340 void bond_alb_deinitialize(struct bonding *bond)
1341 {
1342         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1343
1344         tlb_deinitialize(bond);
1345
1346         if (bond_info->rlb_enabled)
1347                 rlb_deinitialize(bond);
1348 }
1349
1350 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1351 {
1352         struct bonding *bond = netdev_priv(bond_dev);
1353         struct ethhdr *eth_data;
1354         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1355         struct slave *tx_slave = NULL;
1356         static const __be32 ip_bcast = htonl(0xffffffff);
1357         int hash_size = 0;
1358         int do_tx_balance = 1;
1359         u32 hash_index = 0;
1360         const u8 *hash_start = NULL;
1361         struct ipv6hdr *ip6hdr;
1362
1363         skb_reset_mac_header(skb);
1364         eth_data = eth_hdr(skb);
1365
1366         switch (ntohs(skb->protocol)) {
1367         case ETH_P_IP: {
1368                 const struct iphdr *iph = ip_hdr(skb);
1369
1370                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1371                     (iph->daddr == ip_bcast) ||
1372                     (iph->protocol == IPPROTO_IGMP)) {
1373                         do_tx_balance = 0;
1374                         break;
1375                 }
1376                 hash_start = (char *)&(iph->daddr);
1377                 hash_size = sizeof(iph->daddr);
1378         }
1379                 break;
1380         case ETH_P_IPV6:
1381                 /* IPv6 doesn't really use broadcast mac address, but leave
1382                  * that here just in case.
1383                  */
1384                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1385                         do_tx_balance = 0;
1386                         break;
1387                 }
1388
1389                 /* IPv6 uses all-nodes multicast as an equivalent to
1390                  * broadcasts in IPv4.
1391                  */
1392                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1393                         do_tx_balance = 0;
1394                         break;
1395                 }
1396
1397                 /* Additianally, DAD probes should not be tx-balanced as that
1398                  * will lead to false positives for duplicate addresses and
1399                  * prevent address configuration from working.
1400                  */
1401                 ip6hdr = ipv6_hdr(skb);
1402                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1403                         do_tx_balance = 0;
1404                         break;
1405                 }
1406
1407                 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1408                 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1409                 break;
1410         case ETH_P_IPX:
1411                 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1412                         /* something is wrong with this packet */
1413                         do_tx_balance = 0;
1414                         break;
1415                 }
1416
1417                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1418                         /* The only protocol worth balancing in
1419                          * this family since it has an "ARP" like
1420                          * mechanism
1421                          */
1422                         do_tx_balance = 0;
1423                         break;
1424                 }
1425
1426                 hash_start = (char *)eth_data->h_dest;
1427                 hash_size = ETH_ALEN;
1428                 break;
1429         case ETH_P_ARP:
1430                 do_tx_balance = 0;
1431                 if (bond_info->rlb_enabled)
1432                         tx_slave = rlb_arp_xmit(skb, bond);
1433                 break;
1434         default:
1435                 do_tx_balance = 0;
1436                 break;
1437         }
1438
1439         if (do_tx_balance) {
1440                 hash_index = _simple_hash(hash_start, hash_size);
1441                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1442         }
1443
1444         if (!tx_slave) {
1445                 /* unbalanced or unassigned, send through primary */
1446                 tx_slave = rcu_dereference(bond->curr_active_slave);
1447                 bond_info->unbalanced_load += skb->len;
1448         }
1449
1450         if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1451                 if (tx_slave != rcu_dereference(bond->curr_active_slave)) {
1452                         ether_addr_copy(eth_data->h_source,
1453                                         tx_slave->dev->dev_addr);
1454                 }
1455
1456                 bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1457                 goto out;
1458         }
1459
1460         if (tx_slave) {
1461                 _lock_tx_hashtbl(bond);
1462                 __tlb_clear_slave(bond, tx_slave, 0);
1463                 _unlock_tx_hashtbl(bond);
1464         }
1465
1466         /* no suitable interface, frame not sent */
1467         dev_kfree_skb_any(skb);
1468 out:
1469         return NETDEV_TX_OK;
1470 }
1471
1472 void bond_alb_monitor(struct work_struct *work)
1473 {
1474         struct bonding *bond = container_of(work, struct bonding,
1475                                             alb_work.work);
1476         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1477         struct list_head *iter;
1478         struct slave *slave;
1479
1480         if (!bond_has_slaves(bond)) {
1481                 bond_info->tx_rebalance_counter = 0;
1482                 bond_info->lp_counter = 0;
1483                 goto re_arm;
1484         }
1485
1486         rcu_read_lock();
1487
1488         bond_info->tx_rebalance_counter++;
1489         bond_info->lp_counter++;
1490
1491         /* send learning packets */
1492         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1493                 /* change of curr_active_slave involves swapping of mac addresses.
1494                  * in order to avoid this swapping from happening while
1495                  * sending the learning packets, the curr_slave_lock must be held for
1496                  * read.
1497                  */
1498                 read_lock(&bond->curr_slave_lock);
1499
1500                 bond_for_each_slave_rcu(bond, slave, iter)
1501                         alb_send_learning_packets(slave, slave->dev->dev_addr);
1502
1503                 read_unlock(&bond->curr_slave_lock);
1504
1505                 bond_info->lp_counter = 0;
1506         }
1507
1508         /* rebalance tx traffic */
1509         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1510
1511                 read_lock(&bond->curr_slave_lock);
1512
1513                 bond_for_each_slave_rcu(bond, slave, iter) {
1514                         tlb_clear_slave(bond, slave, 1);
1515                         if (slave == bond->curr_active_slave) {
1516                                 SLAVE_TLB_INFO(slave).load =
1517                                         bond_info->unbalanced_load /
1518                                                 BOND_TLB_REBALANCE_INTERVAL;
1519                                 bond_info->unbalanced_load = 0;
1520                         }
1521                 }
1522
1523                 read_unlock(&bond->curr_slave_lock);
1524
1525                 bond_info->tx_rebalance_counter = 0;
1526         }
1527
1528         /* handle rlb stuff */
1529         if (bond_info->rlb_enabled) {
1530                 if (bond_info->primary_is_promisc &&
1531                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1532
1533                         /*
1534                          * dev_set_promiscuity requires rtnl and
1535                          * nothing else.  Avoid race with bond_close.
1536                          */
1537                         rcu_read_unlock();
1538                         if (!rtnl_trylock())
1539                                 goto re_arm;
1540
1541                         bond_info->rlb_promisc_timeout_counter = 0;
1542
1543                         /* If the primary was set to promiscuous mode
1544                          * because a slave was disabled then
1545                          * it can now leave promiscuous mode.
1546                          */
1547                         dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1548                         bond_info->primary_is_promisc = 0;
1549
1550                         rtnl_unlock();
1551                         rcu_read_lock();
1552                 }
1553
1554                 if (bond_info->rlb_rebalance) {
1555                         bond_info->rlb_rebalance = 0;
1556                         rlb_rebalance(bond);
1557                 }
1558
1559                 /* check if clients need updating */
1560                 if (bond_info->rx_ntt) {
1561                         if (bond_info->rlb_update_delay_counter) {
1562                                 --bond_info->rlb_update_delay_counter;
1563                         } else {
1564                                 rlb_update_rx_clients(bond);
1565                                 if (bond_info->rlb_update_retry_counter)
1566                                         --bond_info->rlb_update_retry_counter;
1567                                 else
1568                                         bond_info->rx_ntt = 0;
1569                         }
1570                 }
1571         }
1572         rcu_read_unlock();
1573 re_arm:
1574         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1575 }
1576
1577 /* assumption: called before the slave is attached to the bond
1578  * and not locked by the bond lock
1579  */
1580 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1581 {
1582         int res;
1583
1584         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr);
1585         if (res)
1586                 return res;
1587
1588         res = alb_handle_addr_collision_on_attach(bond, slave);
1589         if (res)
1590                 return res;
1591
1592         tlb_init_slave(slave);
1593
1594         /* order a rebalance ASAP */
1595         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1596
1597         if (bond->alb_info.rlb_enabled)
1598                 bond->alb_info.rlb_rebalance = 1;
1599
1600         return 0;
1601 }
1602
1603 /*
1604  * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1605  * if necessary.
1606  *
1607  * Caller must hold RTNL and no other locks
1608  */
1609 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1610 {
1611         if (bond_has_slaves(bond))
1612                 alb_change_hw_addr_on_detach(bond, slave);
1613
1614         tlb_clear_slave(bond, slave, 0);
1615
1616         if (bond->alb_info.rlb_enabled) {
1617                 bond->alb_info.rx_slave = NULL;
1618                 rlb_clear_slave(bond, slave);
1619         }
1620 }
1621
1622 /* Caller must hold bond lock for read */
1623 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1624 {
1625         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1626
1627         if (link == BOND_LINK_DOWN) {
1628                 tlb_clear_slave(bond, slave, 0);
1629                 if (bond->alb_info.rlb_enabled)
1630                         rlb_clear_slave(bond, slave);
1631         } else if (link == BOND_LINK_UP) {
1632                 /* order a rebalance ASAP */
1633                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1634                 if (bond->alb_info.rlb_enabled) {
1635                         bond->alb_info.rlb_rebalance = 1;
1636                         /* If the updelay module parameter is smaller than the
1637                          * forwarding delay of the switch the rebalance will
1638                          * not work because the rebalance arp replies will
1639                          * not be forwarded to the clients..
1640                          */
1641                 }
1642         }
1643 }
1644
1645 /**
1646  * bond_alb_handle_active_change - assign new curr_active_slave
1647  * @bond: our bonding struct
1648  * @new_slave: new slave to assign
1649  *
1650  * Set the bond->curr_active_slave to @new_slave and handle
1651  * mac address swapping and promiscuity changes as needed.
1652  *
1653  * If new_slave is NULL, caller must hold curr_slave_lock or
1654  * bond->lock for write.
1655  *
1656  * If new_slave is not NULL, caller must hold RTNL, curr_slave_lock
1657  * for write.  Processing here may sleep, so no other locks may be held.
1658  */
1659 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1660         __releases(&bond->curr_slave_lock)
1661         __acquires(&bond->curr_slave_lock)
1662 {
1663         struct slave *swap_slave;
1664
1665         if (bond->curr_active_slave == new_slave)
1666                 return;
1667
1668         if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1669                 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1670                 bond->alb_info.primary_is_promisc = 0;
1671                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1672         }
1673
1674         swap_slave = bond->curr_active_slave;
1675         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1676
1677         if (!new_slave || !bond_has_slaves(bond))
1678                 return;
1679
1680         /* set the new curr_active_slave to the bonds mac address
1681          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1682          */
1683         if (!swap_slave)
1684                 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1685
1686         /*
1687          * Arrange for swap_slave and new_slave to temporarily be
1688          * ignored so we can mess with their MAC addresses without
1689          * fear of interference from transmit activity.
1690          */
1691         if (swap_slave)
1692                 tlb_clear_slave(bond, swap_slave, 1);
1693         tlb_clear_slave(bond, new_slave, 1);
1694
1695         write_unlock_bh(&bond->curr_slave_lock);
1696
1697         ASSERT_RTNL();
1698
1699         /* in TLB mode, the slave might flip down/up with the old dev_addr,
1700          * and thus filter bond->dev_addr's packets, so force bond's mac
1701          */
1702         if (bond->params.mode == BOND_MODE_TLB) {
1703                 struct sockaddr sa;
1704                 u8 tmp_addr[ETH_ALEN];
1705
1706                 ether_addr_copy(tmp_addr, new_slave->dev->dev_addr);
1707
1708                 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1709                 sa.sa_family = bond->dev->type;
1710                 /* we don't care if it can't change its mac, best effort */
1711                 dev_set_mac_address(new_slave->dev, &sa);
1712
1713                 ether_addr_copy(new_slave->dev->dev_addr, tmp_addr);
1714         }
1715
1716         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1717         if (swap_slave) {
1718                 /* swap mac address */
1719                 alb_swap_mac_addr(swap_slave, new_slave);
1720                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1721         } else {
1722                 /* set the new_slave to the bond mac address */
1723                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr);
1724                 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1725         }
1726
1727         write_lock_bh(&bond->curr_slave_lock);
1728 }
1729
1730 /*
1731  * Called with RTNL
1732  */
1733 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1734         __acquires(&bond->lock)
1735         __releases(&bond->lock)
1736 {
1737         struct bonding *bond = netdev_priv(bond_dev);
1738         struct sockaddr *sa = addr;
1739         struct slave *swap_slave;
1740         int res;
1741
1742         if (!is_valid_ether_addr(sa->sa_data))
1743                 return -EADDRNOTAVAIL;
1744
1745         res = alb_set_mac_address(bond, addr);
1746         if (res)
1747                 return res;
1748
1749         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1750
1751         /* If there is no curr_active_slave there is nothing else to do.
1752          * Otherwise we'll need to pass the new address to it and handle
1753          * duplications.
1754          */
1755         if (!bond->curr_active_slave)
1756                 return 0;
1757
1758         swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1759
1760         if (swap_slave) {
1761                 alb_swap_mac_addr(swap_slave, bond->curr_active_slave);
1762                 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1763         } else {
1764                 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr);
1765
1766                 read_lock(&bond->lock);
1767                 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1768                 if (bond->alb_info.rlb_enabled) {
1769                         /* inform clients mac address has changed */
1770                         rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1771                 }
1772                 read_unlock(&bond->lock);
1773         }
1774
1775         return 0;
1776 }
1777
1778 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1779 {
1780         if (bond->alb_info.rlb_enabled)
1781                 rlb_clear_vlan(bond, vlan_id);
1782 }
1783